m625 is not a clinical decision tool · not a publisher · not an expert authority · not a simulator. It is a cross-domain Roll-formalized evidence atlas with public audit + attack surface.
persona: researcher · domain: cancer · tier: 1
ferry target:m590 · fixture id: DEMO-015
self-audit grade:A
i_demo_015 · researcher × cancer · m610 cooked-case re-audit invitation
Why a cancer-research methodologist would come to evidence.x1000.ai
A researcher studying cancer dynamics methodologically — specifically, the question "how does a Roll-formalized cooked case differ from the typical case-report in clinical-research literature" — needs a concrete worked example of the audit pipeline and a path to re-audit. Pain points:
audit standards for methodology papers are weak (no consistent self-audit framework across labs);
cross-paper reproducibility is poor;
attack-surface inventory is rarely public.
m625 does not produce new cancer-research findings. **m625 invites cancer-research methodologists to re-audit m610's 3 base cooked cases under m610's own SELF_AUDIT_BENCHMARK_v0.1 (10 rules × 4 bindings) and m610's MATH_AUDIT pipeline, and to file findings via m625's challenge intake**.
This is not clinical-research advice. This is not an evaluation of any specific paper or sponsor.
The Roll-formalized object: the 3 base cooked cases as audit-replication candidates
(K4.1 bit pattern per m604 V2.6 sovereign reference · Kernel §3.4 · Framework §4.4-4.5; m610 F1-F5 audit families per m604 V2.6 Framework §4.1-4.5.)
97/100 (post-fix · F3 hard-veto fix history is itself the v0.1→v0.2 lesson)
i_case_cancer_450 (R/R DLBCL CD19 CAR-T)
TMU-C-major + TMU-A-minor
(1, 1, 1)phi_direct_rewrite
92/100 manual · 91→98/100 machine
All audit scores are training_data_recall · TODO_unverified (sourced from 01_materials_cache/A3_m610_cooked_cases_anchor.md). Live re-audit is the invitation.
How a researcher methodologist should approach (≤ 5 minutes)
Read the cooked case file. Path: /data/projects/m610_roll_cancer_dynamics/04_cases_cooked/i_case_cancer_001_v0.2.md (read-only from m625's perspective).
Read m610's SELF_AUDIT_BENCHMARK_v0.1. Path: /data/projects/m610_roll_cancer_dynamics/_docs/SELF_AUDIT_BENCHMARK_v0.1.md — 10 rules × 4 bindings (semantic / formal / relational / boundary). This is the audit standard the case was cooked against.
Identify the highest-leverage attack point. For case_150 specifically, the v0.1 → v0.2 transition involved an F3 hard-veto fix (variables in predicate must be V-typed). That fix is the documented audit-break — researchers can attack analogous predicate-variable issues in case_001 and case_450.
File findings. If you find a real attack, file via evidence.x1000.ai/challenge_intake.html (W18) OR via _ops/CHALLENGE_INTAKE.md direct PR. m625 records every fire/no-fire in the append-only public BLOCKED log.
What this demo will not say (anti-drift)
❌ It does not affirm or vouch for the m610 audit scores. The invitation IS re-audit, not validation.
❌ It does not name or rank specific cancer-research labs or principal investigators.
❌ It does not assert that a specific TKI / PARPi / CAR-T pipeline "works"; that is clinical-evidence territory, not methodology.
❌ It does not use cure / reverse / revolutionary / breakthrough lexicon.
❌ It does not imply collaboration with named experts (m517 W7-N inheritance).
Cross-domain transferability declaration (W6.3)
The 3 cooked cases span 3 different K4.1 bit patterns: (0,1,1) cytostatic · (1,1,0) synthetic-lethality · (1,1,1) phi_direct_rewrite. Cross-domain transferability of the audit methodology (not the specific cases) to other domains (neuro · longevity · CKD) is via m620 verify ladder. transferability_pct: bit-pattern necessary, substantive m620 verdict sufficient.
Falsifying prediction
The "m610 cooked-case audit is replicable" claim is falsified by:
an independent re-cook of any of the 3 base cases under m610's MATH_AUDIT_v0.1 pipeline that yields a substantively different K4.1 declaration · OR
an external attack against m610 SELF_AUDIT_BENCHMARK_v0.1 rules that demonstrates spec ↔ code ↔ test drift.
AS-021 (partial truth without falsifiability) — G3.3 + G2.2
Ferry target
m590 (claim_pack · for graph anchoring) · m620 (verify_req if researcher requests live re-audit).
Honest verdict (W6.5)
已做: researcher × cancer audit-replication invitation · 3 base cases summarized · audit-replication path documented · public attack-intake established.
待办: pre-recorded re-audit example (v1.0+ · operator runs python3 -m 02_audit.audit on m610 cooked files and publishes the result).
漏洞: m610 audit scores are training_data_recall · TODO_unverified from the cache · live verification is m610 sovereign work; v0.2 m610 v0.6+ release may differ.
反驳预案:
- *"This is just an invitation, not evidence."* — Correct. The framework's value is the audit-replication procedure, not new evidence.
- *"What if no researcher takes you up on this?"* — Then the framework correctly remains untested by external audit. v0.3+ public BLOCKED log will reflect that honestly.